INVESTIGATING DATA GENERATION STRATEGIES FOR LEARNING HEURISTIC FUNCTIONS IN CLASSICAL PLANNING

In classical planning, the efficiency and effectiveness of heuristic functions are crucial for guiding search algorithms toward optimal solutions. This study investigates various data generation strategies for training machine learning models to learn heuristic functions in classical planning domains. By comparing approaches such as random sampling, goal-directed sampling, and domain-specific guided data collection, the research evaluates their impact on the accuracy and generalizability of learned heuristics. Experimental results across benchmark planning problems reveal that the choice of data generation strategy significantly influences the performance of the resulting heuristics. The study provides insights into the trade-offs between data diversity, representativeness, and computational efficiency, contributing to the development of more robust learning-based planning systems.